OpenAI is coming for the app layer
Also: Nvidia, the bellwether of our times
Welcome to Cautious Optimism, a newsletter on tech, business and power.
📈 Trending Up: You don’t say … May I meet you? … money buys weapons … bone saws … ICE raids in the south … Alphabet shares … GDP growth in the EU … Apptronik
📉 Trending Down: Trade deal popularity … kids using AI … China-Japan relations … the cost of losing weight … POTUS popularity … data center approval … Americans’ purchasing power … the Chinese housing market … Japanese GDP
Things That Matter
Startup of the day: Flatpay. The old riff that goes, “your margin is my opportunity,” is one of my favorite bits about tech. Since this industry, by definition, allows for greater gross profitability, companies compete on price and the economy improves.
Flatpay is a good example of this phenomenon. While Square will charge you around 2.5% and $0.15 per swiped card, and Stripe wants 2.9% and $0.30 per “successful transaction [on] domestic cards.” Flatpay is going lower.
It charges between 0.99% and 1.49% for both online and offline payments, with a focus on simplicity. Small businesses love this point-of-sale company so much that it’s soaring:
“We crossed €100 million of ARR in October,” Janca-Jensen told TechCrunch. He added that this amount (approximately $116 million) is increasing by nearly €1 million a day ($1.16 million). “The plan for 2026 is to grow another 300%, so hopefully leave the year with between €400 and €500 million of ARR.”
Presumably, the ARR mentioned here is annualized run rate and not annual recurring revenue, as Flatpay makes more from its usage-based fees than anything else. Regardless, it’s an impressive set of numbers. The startup just raised new capital at a $1.75 billion valuation, so by the end of 2026, it’s either going to be worth a lot more or it will become one of the most insane bargains on the market.
Good on the Danish startup for attacking incumbents with lower pricing. That’s precisely what we want to see from a tech startup. So much for writing Europe off, yeah?
The bellwether of our times: This Wednesday, Nvidia will report third-quarter results. Analysts expect revenue of $54.8 billion (+56% YoY) and profit of $1.25 per share, but the company’s forecast will prove the most important.
Nvidia CEO Jensen Huang said in October that his company already has $500 billion worth of orders for its AI gear for 2025 and 2026. If that’s true, the company’s growth curve could grow even steeper next year. Investors will want to know specifics regarding demand.
The enthusiasm and hype around AI has morphed to concern and pessimism, simply because the AI investments we’re seeing are world-bending, AI revenues are nascent in comparison, and it seems model progress has slowed.
Nvidia could help ameliorate worries, but only some. The better the company’s Q3 results and forecast are, the more the AI ‘bubble’ may inflate. Still, it’s better if Nvidia crushes.
Two major Nvidia shareholders have cashed in their holdings in recent weeks: SoftBank and Peter Thiel. Taking profits is no sin, but big exits like these just ahead of a company’s quarterly results are notable.
Jeff got bored: Apart from celebrating his space company’s recent booster landing, Amazon founder Jeff Bezos is working on Project Prometheus. The startup has raised a staggering $6.2 billion, and features the former Amazon boss as co-founder and co-CEO alongside Vik Bajaj, who co-founded and led Google’s life sciences unit, and also founded Foresite Labs.
What will Project Prometheus do? Per the Times, it’s “focusing on technology that dovetails with Mr. Bezos’ interest in taking people to outer space,” by working to build AI that “will help in engineering and manufacturing in a number of fields, including computers, aerospace and automobiles.”
Given that Bezos has investments in computers (AWS), aerospace (Blue Origin) and automobiles (via his investment in Slate), one can see the synergies.
Forget private jets, sprawling yachts, or merely owning your own space company. Real billionaires these days fight to control their own nation-state-sized empire of companies. Double the points if the collection is synergistic, akin to Musk merging X and xAI, or Bezos investing in AI that could help his other efforts.
Crypto is getting hammered: The IPO class of 2025 included a host of crypto-friendly companies like stablecoin giant Circle, crypto exchanges Gemini and Bullish, blockchain lender Figure, and consumer trading services with crypto elements like eToro. The crypto up-cycle was driven in part by rising token values, especially Ethereum, whose token rose to around $4,300, and Bitcoin, which reached about $118,000.
Today, you can snag a full bitcoin for $94,900, and ether’s fallen to $3,160. Bitcoin fell 10.6% over the last week, while Ethereum’s token declined 12.2%. Solana is down 16.8% as well.
Crypto trading volumes are linked to volatility and rising prices. Falling token values is crypto trading poison. That’s why such declines tend to delete revenue for both public and private crypto exchanges.
Why the declines, though? Economic concerns impact the most speculative investments first, and often more harshly than their more boring counterparts. And despite their growing institutional maturity, crypto is still pretty darn nascent.
OpenAI is coming for the app layer
AI is a roughly 2.5 horse race at the moment. China’s AI model companies (DeepSeek, Moonshot, etc.) are competing with America’s AI model companies (OpenAI, Anthropic, xAI, etc.), which in turn are competing with Europe’s smaller group (Mistral, Black Forest Labs, etc.). Japan wants to join the club, too, and its rising national champion Sakana AI just raised $135 million:
“We plan to deploy the new capital for not only R&D, including [AI] model development, but also to expand our workforce for engineering, sales, and distribution teams in Japan,” [CEO David Ha] said. This year, Sakana has expanded its presence in Japan, teaming up with leading local enterprises, including Daiwa and MUFG, and developing AI solutions.
Foundation models aren’t the only game in town, of course. Voize (medical AI note-taking) just raised a $50 million Series A, and Turing’s self-driving tech landed $99 million, giving Japanese AI’s ecosystem another win to start the week.
Here’s a question we haven’t asked in a minute: Will startups building on top of AI models generate more value than the AI model companies themselves, or will Big Model crush the AI app layer?
Many VCs have split their bets, throwing cash at the startups that use AI models from the major labs while also backing the labs themselves. Heads they win, tails they win. You could even argue that the investors active on both sides of the model-app layer question will be right twice.
Here’s OpenAI’s Applications CEO Fidji Simo in conversation with Wired about her company’s financial health, and plans to grow into its spending:
OpenAI is obviously one of the world’s most valuable startups, if not the most valuable, but it’s also losing billions of dollars every year.
I’ve noticed.
What opportunities do you see to get it on a path to profitability?
It all comes back to the size of the markets and the value we’re providing in each market. In the past, only the wealthy had access to a team of helpers. With ChatGPT we could give everyone that team—a personal shopper, a travel agent, a financial adviser, a health coach. That is incredibly valuable, and we have barely scratched the surface. If we build that, I assume that people are going to want to pay a lot of money for that, and that revenue is going to come.
Meanwhile, on the enterprise side, we sell an API and ChatGPT Enterprise, which is a great product but a very thin layer compared to all the things that we could be building for enterprise. If you think about building agents for every industry and function, there is so much to build, either by us or by enabling third parties to build on top of our platform.
So I’m like, OK, the markets are huge. The depth of value is huge. That’s the basic formula for monetizing. Then the real question becomes, will we have the compute to deliver that?
It sounds like OpenAI intends to win the consumer outright while working with startups and other companies to dominate the enterprise side. From its current position as the default consumer AI company today, the perspective makes good sense; why not build on top of what you are already leading?
For consumer AI startups, the fact that OpenAI intends to go ham on monetizing the average AI user is probably bad news. Business software AI companies aren’t safe, either. Anthropic wants to own the B2B AI market. So you have two giants coming for both sides of the model-app layer divide.
Should startups pack up and hit the bricks? No, but I think that the more niche a startup’s AI use case, the better the chances it will avoid getting stepped on by a giant. Conversely, the more general your offering is, the greater the chances that OpenAI or Anthropic step in to absorb your proven revenue generation model. Anything that is a known win will become part of the ‘borg.
If the market plays out that way, AI startups would find themselves more often playing the role of pioneers than masters of the largest market categories. It’s a smaller prize up front but a potentially lucrative business model over time.
But with VCs expecting ever-faster growth, is it a viable one? Probably not. So we’re going to see startups chase popular categories (notes, transcription, translation, coding, legal AI help, etc.) in hopes of getting so far ahead of the AI model labs as to become unbeatable, or at the very least become attractive M&A targets.
OpenAI needs every dollar of revenue it can get its hands on. That means it’s going to play for keeps everywhere, all at once.